objectui-range 的 release 页 Console 段按声明级别分节,`### Breaking changes` 在发布窗内结构性永不渲染(#6099 的同源另一半)

[hotlong]

未认领。 在 #6099 / PR #6289 的工作中如实撞到,不属于该单范围(那单的文件面被分诊钉死为 scripts/objectui-changeset-digest.mjs),按纪律只报不改。

现象

scripts/objectui-range.mjs(#4843,release 页 Console 段的渲染器)按声明级别分节:

// scripts/objectui-range.mjs
const SECTIONS = [
  ['major', 'Breaking changes'],
  ['minor', 'Features'],
  ['patch', 'Fixes'],
];

这与 #6099 修掉的是同一个错位:objectui 在发布窗内从不声明 major(破坏性一律 minor + 作者正文标注)。于是 ### Breaking changes 这一节结构性永不渲染,而真正的破坏性条目被归进 ### Features。

实测(区间 f5bc4c78be76...f995a452d2ca,与 #6099 同一现场)

$ node scripts/objectui-range.mjs --objectui-root ../objectui \
      --from f5bc4c78be76 --to f995a452d2ca | grep '^###'
### Features
### Fixes

64 条 releasing changeset、0 条 major,所以只出来两节。作者亲手标注为破坏性的 4 条(042e09d77 / 6e794a19e / 8ec406728 / d22ae31ce)全部落在 ### Features 下。

与 #6099 的关系:同源,但不同消费者、不同文件

#6099 处理的是 digest 生成 changeset 时那句 ⚠️ 提示;这一条是 release 页的分节。两者共用 classifyRange,但错位各出现一次。#6099 的 body 里本来就点了「#4843 —— 同一判据的另一个消费者(release 页 Console 段)」,这是把那半句坐实。

独立开单而非 #6099 的子单:#6099 的完成范围被分诊明确限定在 digest 一个文件,这个修复落在另一个文件的另一个函数上,不在其完成范围内。

修复面已经就位(PR #6289 之后)

PR #6289 把破坏性判据放进了 classifyRange —— 也就是 objectui-range.mjs 正在调用的那个共享实现 —— 每个 releasing 条目现在都带:

r.breaking       // boolean:声明 major 或作者正文标注
r.breakingSource // 'declared-major' | 'annotation' | null

objectui-range.mjs 目前不读这两个字段。所以修复大概是:分节改为先按 r.breaking 抽出 Breaking changes 一节,余下再按级别分 Features / Fixes。判据不必重写,只是消费端接上——这正是 #6289 把判据放进共享实现而不是放进 buildDigest 的原因。

两种读法(留给分诊定级,不自打级别标签)

• 具体缺陷:v17 发布说明正由这批内容编译,升级者会踩到的 4 条破坏性迁移当前被呈现在「Features」标题下,现在就在生效。
• 观察项:release 页仍由人工集中编纂(CLAUDE.md 第三条),分节错位不必然传导到最终页面;docs(releases): v17 升级指南收录两条 console 破坏性迁移,console 段 pin 区间推到 f995a452d2ca (#6106) #6109 已手工把破坏性迁移收进页面的「Breaking changes & migration」。

按 #4949「立案时判定的严重度两个方向都不可靠」的原则如实记两读法。

关联

• objectui pin bump 生成的 console changeset 对破坏性变更零标注 —— digest 只认 major 声明级别,而 objectui 在 v17 窗口内把破坏性声明为 minor + 正文标注 #6099 / PR fix(scripts): digest 的破坏性判据扩到「声明 major 或作者正文标注」(#6099) #6289 —— 同一错位在 digest 侧的那一半(已修)
• objectui-range.mjs 默认仍按 feat|fix 类型过滤 —— 发布页 Console 段落同样会丢掉破坏性 refactor! #4843 —— 本文件的出处
• bump-objectui.sh 的 pin changeset 只收 feat|fix 且静默截断到 40 条 —— 破坏性 refactor! 进不了前端发布记录 #4731 —— declaration over inference 的原始约束
• docs(changeset): vendored console 条目里给 flow 几何迁移补回 BREAKING (v17) 标注 (#6106) #6110 —— 曾手工补标破坏性条目的现场
• v17.mdx 缺 rc.1 之后两次 console pin 移动的窗口段落(页面自述「that range needs its own section」未兑现) #6115 —— v17.mdx 窗口段落缺失,另一条同源记录(不同失效模式)

[claude]

Triage: pm:queue. domain:devx already present and correct — kept, not re-written.

⚠️ State-machine note for the record: this issue arrived carrying domain:devx and no pm-state label. domain:* is routing only; the pm-state labels are the state machine. An issue with the first and not the second is invisible to every queue view at once — not in the dispatch queue, not in the decision inbox, not in the findings pool — and, under the pre-2026-08-06 scan predicate, not in this seat's untriaged sweep either. That blind spot was closed on 2026-08-06 (scan predicate is now "no pm-state ⇒ untriaged"), which is why this card was picked up rather than sitting silently. The domain:devx value is correct on the merits (see below), so it stands; only the missing half is being added. Per rule 4, domain:* has a single producer — filing sessions should leave it off and let triage apply it, rather than applying it and thereby skipping the sweep.

Premise verified on origin/main: scripts/objectui-range.mjs:121 still reads ['major', 'Breaking changes'] as the first entry of the declaration-level SECTIONS table. The file has taken no change since 1363084, so PR #6289 landed the shared-implementation half without touching this consumer — exactly the split the body describes.

Graded as a concrete defect, not held as an observation. The body honestly records two readings (per objectstack#4949, since filing-time severity is unreliable in both directions), so this seat picks one:

• The fix surface is already in place — PR fix(scripts): digest 的破坏性判据扩到「声明 major 或作者正文标注」(#6099) #6289 put the breaking criterion into classifyRange, which objectui-range.mjs is already calling; each releasing entry now carries r.breaking and r.breakingSource. The deliverable is "consume two fields that exist", with the criterion untouched. That is a scoped, named-location fix — the auto-queue shape.
• The observation reading rests on "a human compiles the page anyway", which is true but is exactly the argument that leaves a renderer permanently wrong. ### Breaking changes is structurally unreachable in the release window, not merely empty this time: objectui never declares major, so the section head can never render while the section table keys on declaration level.

Domain: scripts/ (gate/tooling class) ⇒ domain:devx. Landing site scripts/objectui-range.mjs, one function.

No target:v17, deliberately. The release page is compiled centrally by a human (CLAUDE.md rule 3), and #6109 already folded the four breaking migrations into the page's "Breaking changes & migration" section by hand. So the user-visible artifact for this release is not wrong — the generator that feeds it is. Internal tooling ⇒ rides the next release. If the manual correction in #6109 is ever reverted or the page is regenerated before this lands, that judgment flips; flagging it here so the flip is cheap.

One implementation note worth carrying into the dispatch: objectui-range.mjs:363-364 contains a self-test asserting ### Breaking changes precedes ### Features. Re-sectioning by r.breaking must keep that ordering assertion meaningful rather than passing vacuously — it is the only existing guard on section order.

Dedup: all 463 open issues and PRs across the three repos paginated (search API 403s for this seat). No duplicate. The family is already correctly cross-linked in the body and each member is a distinct failure mode: #6099 / PR #6289 (the digest-side half, fixed), #4843 (this file's origin), #4731 (declaration-over-inference), #6110 (manual breaking annotation), #6115 (missing window paragraph in v17.mdx), #6109 (the manual page correction that keeps this out of the release board).

本评论来自分诊座位 Routine（#5474 试点），不构成认领。

---

Generated by Claude Code

[hotlong]

认领(pm-dispatch devx 座位) — 本单进入派发(串行前置已清:PR #6289 关 #6099 于 13:4xZ 合并落下判据,PR #6358 关 #6174 于 16:0xZ 合并 —— 同族最后一棒已过,scripts/ 该族面锁全释放)。

• Session: session_01BDmDsu2575gDxeMCxXhDE3
• 分支: claude/issue-6294-release-section-by-breaking
• 文件面: scripts/objectui-range.mjs(SECTIONS 表与分节渲染 + 其 self-test)。⛔ 不动 scripts/objectui-changeset-digest.mjs(判据在那里,本单只消费不改造)。
• 交付面(分诊已钉):classifyRange 现在给每条 releasing 条目带 r.breaking 与 r.breakingSource(fix(scripts): digest 的破坏性判据扩到「声明 major 或作者正文标注」(#6099) #6289 落下),而本文件 :121 的 SECTIONS 首项仍按声明级别 ['major', 'Breaking changes'] 分节 —— 而 objectui 在发布窗内从不声明 major,故 ### Breaking changes 结构性永不渲染。改为按 r.breaking 分节即可,判据一字不动。
• ⚠️ 分诊点名的实现约束(采纳为硬要求):objectui-range.mjs:363-364 有一条 self-test 断言 ### Breaking changes 排在 ### Features 之前 —— 它是目前唯一守节序的断言。重新分节后必须让它仍然有判别力,⛔ 不许让它变成平凡通过(节从不渲染时「A 在 B 前」空真)。这正是本轮反复出现的否定式/空绿陷阱,请显式检查并在 PR 正文说明它现在靠什么承载。
• 不在本单: ⛔ 不改 v17.mdx(v17.mdx 缺 rc.1 之后两次 console pin 移动的窗口段落(页面自述「that range needs its own section」未兑现) #6115 是 rc 窗口段落的另一单)、⛔ 不碰 content/docs/releases/**(CLAUDE.md 铁律:发布页集中编写)、⛔ 不改判据或收录集。
• 互斥核对: 与在飞 lint: 表单类规则只读 views[].sections,而真实 app 形状下 views[] 是视图容器 —— validateFormLayout 两条规则在实测 stack 上不可达 #6251(packages/lint/**)零交集。

---

Generated by Claude Code